1

DISCONTINUITIES, CONVERGENCE AND SURVIVAL OFINEFFICIENT TRAJECTORIES IN TECHNICAL PROGRESS

Andrea Bonaccorsi, Paola Giuri, Francesca Pierotti

Sant’Anna School of Advanced StudiesVia Carducci , 4056127, Pisa – Italy

Phone: +39.050.883359, Fax: +39.050.883343e-mail: bonaccorsi@sssup.it, giuri@sssup.it, pierotti@sssup.it

First draft

September 2001

2

1. Introduction

The paper discusses some central issues in the evolutionary analysis of technological progress and

of technological competition among actors. It also develops a methodological contribution by building

a measure of technology based on multiple technical parameters.

The paper presents an empirical analysis of the technological progress in the commercial aero-

engine industry since 1948 to 1997, based on two proprietary databases. The Atlas Aviation Database

contains all the transactions occurring from 1948 to 1997 between engine manufacturers, aircraft

manufacturers and airline companies in the market for large commercial aircraft. For each transaction

the aircraft and engine product version is specified. The AirTech database contains 16 technical

parameters for each 114 engine version.

We develop a multidimensional measure of technical progress through Data Envelopment Analysis

(DEA). This mathematical linear programming method builds an envelop of a number of inputs for

maximising (minimising) one or multiple outputs. The choice of key technical parameters of aero-

engines (SFC, BPR, OPR and Thrust/Weight) and the definition of inputs and outputs has been

carried out with the support of aeronautical engineers and purchasing managers of a major airline

company. The approach provides for each engine a measure of efficiency based on the contribution of

each technical parameter in the overall design of the engine.

The observation over time of the multi-dimensional measure of efficiency shows an increasing

pattern evolving through different trajectories. The empirical analysis allow to identify some structural

patterns in the technological evolution at the industry the firms and at the firm level.

Some main conclusions are drawn. Technological discontinuities observed at the level of single

technical parameters are much more marked than for the multi-dimensional measure obtained through

DEA. Technological progress is very discontinuous only with respect to a single parameters while it is

more gradual with respect to the bundle of performance indicators linked by complementary relations.

Moreover, discontinuities does not only bring about convergence in the space of technical parameters

but may open multiple equilibria which allow survival of inefficient products.

The paper develops a discussion of relevant issues in the evolutionary debate on technology and

present detailed empirical analysis and cases at the firm level showing common strategies of

convergence on the frontier and differentiated strategies of introduction of products below the frontier.

2. Data

The objects of the empirical analysis is the commercial jet aero-engine industries since its birth,

from 1958 to 1997. The choice of the aero-engine industry has a number of reasons:

3

- well-defined brands and generation of products;

- availability of complete directory of data;

- existence of a stable set of technical and performance characteristics representing the product

over the history of the industry.

The empirical analysis is based on two databases.

The Atlas Aviation Database contains all transactions (orders) occurring from 1948 to 1997 between

engine manufacturers, aircraft manufacturers and airline companies in the market for commercial jet

and turboprop aircraft. For each transaction the aircraft and engine product version is specified.

The AirTech database contains 16 technical parameters for 114 jet engine versions and 11 technical

parameters for 76 turboprop engine versions and two basic parameters for each aircraft included in the

Atlas database (Table 1). The AirTech database has been built by using several sources of data: Jane’s

All the World Aircraft 1950-1998, Jane’s Aero-engines 1997, Flight International 1970-2000, Aviation Week and

Space Technology 1970-2000, engine and aircraft companies web sites, company reports, product

brochures, technical data provided by a major airline company, phone contacts with technical and

information offices of two of the larger aero-engine companies.

Further information and details have been drawn from publications on the history of the aviation

technology and on the structure of the aircraft and aero-engine industries (among others Miller and

Sawers, 1968; Phillips, 1971; Constant, 1980; Mowery and Rosenberg, 1982; Vincenti, 1990; Garvin,

1998).

Technical parameters have been classified in technical characteristics and technical performance

parameters, or service characteristics with the terminology of Saviotti and Metcalfe (1984).

In a previous paper (Bonaccorsi and Giuri, 2001) we studied the evolution of single technical

performance parameters and the technological position of companies with respect to the frontier. The

analysis has been carried out at the industry level and at the sub-market level. Sub-markets have been

identified through cluster analysis, which classified aero-engines according to 5 selected technical

characteristics (weight, length, diameter, thrust, airflow). Cluster 1 represents the smaller segment of the

market, including the first turbojets introduced at the birth of the industry, the turbofans introduced at

the beginning of the 1960’s in substitution of the turbojets, the second generation of turbofans and the

small regional jets of the 1990’s. Cluster 2 includes larger size-engines while cluster 3 includes the

largest engines of the three big players, which power very large aircraft.

The choice of technical parameters and the distinction in characteristics and performance

parameters has been validated through interviews with aeronautical engineers and with managers of the

purchasing division of a major airline company1.

1 Some of them, like thrust, can be difficult to classify because they can be considered technical or performancecharacteristics, depending on the level of education of customers.

4

Table 1. List of product characteristics

Type of characteristics CharacteristicsTechnical CompressorTechnical Engine type (output)Technical Combustor typeTechnical N° fansTechnical N° LP compressorsTechnical N° HP compressorsTechnical N° of turbinesTechnical N° HP turbinesTechnical Air flow - lb/secTechnical Length – inchTechnical Diameter – inchTechnical Weight-dry - lbTechnical Thrust TO – lbPerformance BPR (By pass ratio)Performance OPR (Overall pressure ratio)Performance SFC (Specific Fuel Consumption) TO - lb/hr/lb

3. Building multi-dimensional measures of technology

3.1 Data Envelopment Analysis

To develop a multi-dimensional measure of technology we refer to previous contributions on

production and cost frontiers (Farrell, 1957), which study the efficiency in combining a set of inputs to

obtain a set of outputs.

Before Farrell’s contributions in 1957, the empirical literature on the estimate of production and

cost frontiers used least squares regression models to explain the relation between a set of inputs and

an average level of output. The emphasis was on the average production technology rather than the best

technology. Moreover, no explanation was provided for the significance and the sign of the residuals of

the regression. In Farrell (1957) the definition of the production frontier is based on the best

technology, and the study of the efficiency is based on the concept of distance from the frontier.

This issue has been developed within an econometric and a mathematic approach. The first is a

stochastic and parametric approach and the second is deterministic and non parametric. They differ

mainly in two points: the determination of the shape and position of the frontier, and the interpretation

of the distance from the frontier, that is the measure of efficiency.

In this work we use the mathematical programming approach called Data Envelopment Analysis

(DEA), which has been originally proposed by Farrell (1957) and subsequently developed by other

5

authors (Charnes e Cooper, 1985; Knox Lovell and Schmidt, 1988; Norman and Stoker, 1993;

Grosskopf, 1993). Its basic characteristics are that it does not require the specification of a functional

form with the introduction of parameters, and the combination of inputs and outputs is deterministic,

because of the absence of an error term.

The method builds a production frontier by solving a number of mathematical programming

problems, and computes a measure of technical and allocative efficiency for each production unit.

Technical efficiency is the maximum output given a set of fixed observable inputs (output oriented), or

the minimum level of inputs given an observable level of output (input oriented).

The DEA is deployed in three main steps:

− construction of the set of inputs and outputs;

− computation of the Farrell technical efficiency measure;

− computation of the minimum cost for each production unit and of allocative efficiency.

For the purposes of building a pure technological frontier, we refer to measures of technical

efficiency, and not of allocative efficiency, which involve the use of price for their computation.

The technical efficiency (input oriented) is defined as

Fi(y,x C,S)=min{λ:λx∈ L(y C,S)},

subject to:

1. mkm

K

kk yyz ≥∑

=1

2. nkn

K

kk xxz ≤∑

=1

3. 0≥kz

in the case of constant return to scale (C)2 and strong disposability of inputs (S)3, where:

nx is the set of n=1,2,…,N inputs,

2 The model can be built on three different assumptions on returns to scale:constant returns to scale: 0≥kz k∀ ;

non increasing returns to scale : 11

≤∑=

K

kkz ;

variable return to scale: 11

=∑=

K

kkz .

3 The set of inputs and outputs may have the following properties:strong disposability of inputs: if inputs are constant or decreasing, output is non decreasing;weak disposability of inputs: if inputs increase proportionally, output is non decreasing;strong disposability of outputs: it is possible to freely dispose of outputs (typical assumption about disposability of outputs);weak disposability of outputs: proportional reductions of all outputs are feasible. However, it does not necessarily follow thatreductions in individual outputs are feasible (i.e. the reduction of undesirable outputs may be costly).

6

my is the set of m=1,2, …, M outputs,

kz is the set of intensities of n inputs and m outputs for k=1,2,…,K units of productions,

Fi is the measure of input oriented technical efficiency,

L is the Input Requirement Set, which defines the assumptions on returns to scale and substitutability

of inputs.

A firm is technically efficient if Fi=1, is technically inefficient if 0<Fi<1.

The advantages of DEA are the possibility of using at the same time a large number of inputs and

outputs, and of obtaining measures of efficiency for both the set of inputs and outputs. Moreover, this

approach is not influenced by problems of mispecification of error and functional form.

The main limitation is that every deviation from the frontier is considered as inefficiency, and it is

not possible to statistically estimate the deviation from the frontier. As a consequence, it is very

important to reduce as much as possible errors of data measurement and of variable selection.

Recent studies are trying to overcome these limitations, through the development of non parametric

stochastic frontiers and stochastic mathematical programming approaches.

3.2 Technological frontier in the aero-engine industry

By applying DEA, we develop a technological frontier, which is composed by multiple technical

parameters.

The first step is the definition of variables that represent the inputs and outputs in the DEA.

The selection of the technical parameters to be used for the definition of a multi-parameter

technological indicator has been done with the support of aeronautical engineers and trough the

consultation of technical literature on the structure and functioning of aero-engines (Janes Aero-

engines, 1997; libri…)

A good measure of efficiency (distance from the frontier) can be well obtained by combining the

following 4 main technical performance parameters, which embody a number of other technical

parameters and interactions among them:

− Specific fuel consumption (SFC),

− By-pass ratio (BPR),

− Overall pressure ratio (OPR)

− Trust/Weight (T/W).

The first three are performance parameters, as indicated in Table 1; T/W is a performance indicator

built by the ratio of the two technical parameters thrust and weight. Advancements in these parameters

depend on innovation in other technical parameters and design characteristics.

7

The efficiency with which the turbojet converts fuel energy into useful propulsive power is usually

expressed in terms of specific fuel consumption. This is defined as the mass flow rate of fuel burned

per unit of net thrust produced (Janes Aero-engines, 1997).

For the application of DEA we defined SFC as the output of the process, and BPR, OPR and T/W

as inputs, which are all related to the output SFC.

BPR is “the ratio of the air flow bypassing the core of the engine to the air flow entering the core”

(Janes Aero-engines, 1997).

OPR is “the ratio of the outlet pressure from the compression process to the inlet pressure to the

compression process” (Janes Aero-engines, 1997).

Thrust is a design parameter, which is directly related to the weight of the engine. The intensity of

the relation is a performance parameter. In fact for having an increase in thrust, the weight of the

engine also increases, and this reduces efficiency. The ratio T/W can keep constant only if there is

technological innovation.

The inputs and outputs are related in the following way.

SFC is influenced by the propulsive efficiency expressed by BPR. The propulsive efficiency is the

ratio of the useful propulsive power applied to the aircraft to the kinetic energy of the gas leaving the

nozzle. There is an inverse causal relation between BPR and SFC. At the same time a high BPR

increases weight and can negatively affects OPR.

SFC is also inversely influenced by the thermal efficiency, which is “the ratio of the kinetic energy of

the gas leaving the nozzle to the fuel energy” which depends on the OPR.

There is also a functional relation between T/W and SFC. However the relation between the two

parameters is not direct, but it is mediated by other technical. For having technological advance, the

direction of the relation has to be negative.

The second step is the computation of the level of efficiency of each aero-engine, which is carried

out by using the software OnFront 2 (Fare and Grosskopf, 2000), designed for efficiency and

productivity measurement.

For the purposes of this study, the suitable approach is the output oriented. However, in OnFront

the output oriented approach maximises output given a set of a fixed observable inputs. Our objective

is the minimisation of output (specific fuel consumption) given a level of inputs. To solve a

computational problem due to the lack of the procedure in the software4, we use the input oriented

approach (minimising inputs given a level of outputs) considering the output SFC as input to be

minimised and the inputs OPR, BPR and T/W as given outputs.

4 The software allows minimisation of outputs for determining allocative efficiency in a cost frontier. However, theoptimisation problem of the cost frontier, which requires the use of price data, is different from the objective ofdetermination of a pure technological frontier.

8

We computed the efficiency under the assumptions of constant returns to scale and strong

disposability of inputs and outputs. We also computed the efficiency under the other assumptions

defined above. The results were highly correlated, and we presented the results with the standard

assumptions.

The result of the application of DEA is an index of efficiency for each aero-engine in a scale 0-1. An

index close to 1 indicates a high level of efficiency in the combination of inputs and outputs, an index

close to 0 indicates a low level of efficiency. Notes that 1 represents the best combination of inputs and

outputs among the 114 aero-engines analysed. The efficiency index for each engine is presented in

Table 2.

Next section will discuss the results of the empirical analysis.

3.3 Technological progress

The rate of technical progress is determined by calculating 3-year rates of change of the level of

efficiency over time. The rate of change of the level of efficiency is compared with the rate of change

of the single performance parameters to show the presence of major discontinuities in technical

progress.

The values of single parameters are normalised on a 0-1 scale to allow comparability with the

efficiency measure.

The rate of change is computed as follows

1max

1maxmax

−

−−= t

k

tk

tkt

k KKKT

where t = last year of 3-year periods

k = 1, 2, …z performance indicators

1max−t

kK is the best value of k at time t0 across all products in the industry.

4. Empirical analysis and discussion

The rich array of data makes it possible to offer an interpretation of the long-term evolution of the

jet engine technology both at the level of the entire industry and of individual manufacturers. We focus

here on some of the crucial points in the debate on the evolutionary theory of technology, for which

we can offer a fresh contribution.

9

4.1 Gradualism vs. punctuation in the evolutionary process

Figure 1 shows the maximum level of the DEA indicator of each company for each year of the time

series across all the industry. Figures 2-4 replicate the representation for the three industry clusters

defined above.

The technological evolution is one of sustained progress over a very long period, with remarkable

improvements in overall efficiency. The level of the DEA index changes from around 0.2 in the 1950s

to a value of 1 in the 1990s.

In the initial period of the history of the industry, the average level of efficiency is around 0.2, a

very low level in comparison with the subsequent technological evolution. The cloud of points in

Figure 1 include the Avon models based on the initial turbojet technology and the Conway models,

with a by-pass ratio of 0.33, and the JT3C and JT4A turbojet engines of Pratt & Whitney. Then there is

a first jump to levels of DEA around 0.4 - 0.5 in the following ten years: these engine models

incorporate the new turbofan technology and are developed by Rolls Royce (the Spey family) and Pratt

& Whitney (the JT3D family). These two families do not undergo any significant change over time in

terms of their overall efficiency.

The true radical change in the jet engine technology comes in with the new concept of high by-pass

ratio. Models introduced in the early 1960s realise a jump of the DEA index from 0.5 to 0.7, and, more

importantly, open a new trajectory that allows a series of further improvements.

Finally, a further jump is visible in the 1990s with the introduction of the very large engines for the

B777 (GE90-92, P&W 4074, 4077 and 4090, the RR Trent family).

The overall pattern is one of three discontinuities in overall efficiency, followed by long periods of

incremental advancements.

The graphical representation disaggregated at the cluster level (Figures 2-4) confirms the presence

of three big jumps in cluster 1. However, in cluster 1 the third important change occurs later than in

other clusters, except for CFM International, which enters introducing innovative products with an

efficiency level higher than 0.7. Cluster 2 includes all larger families of engines of the third generation

with a DEA ranging from 0.6 to 0.9. Cluster 3 shows the jump to DEA indexes of 1, realised by the

engines for the larger aircraft (B747, B777).

Does this result support the notion of punctuated equilibrium in evolutionary processes?

Figure 5 shows the evolution of rates of change of single technical parameters and DEA efficiency

in the history of the technology in terms of rates of change on a 3--year basis. A clear discontinuity

pattern in single parameters emerges, with extremely high rates of change. When looking at the

dynamics of the DEA measure, however, rates of change are much lower. There is no sharp

discontinuity in the data, but rather an overall upward trend, with some points of acceleration. The

10

overall picture is one of strong discontinuity in single parameters and less discontinuity in aggregate measures of

technical performance.

This finding has two important implications for evolutionary theory.

From a methodological point of view, it must be remarked that most of the discussions on the

nature of technological progress, namely, the continuous vs. discontinuous or punctuated nature, is

based on the observation of time series of single technological indicators, usually referring to a particular

technological performance of a product (Sahal, 1985a; Tushman and Anderson, 1986, 1990; Levinthal,

1998). In most cases, this is due to practical limitations in the availability of data for long time series. In

some sense, nobody denies that the very concept of technological performance of a product is a truly

multidimensional one, but it becomes very difficult to trace many variables over a sufficiently long

period to allow significant analysis. So there is a sharp separation between methodologies that use a

large array of variables, but with a limited time horizon, such as hedonic price in microeconomics or

conjoint analysis in marketing science, and methodologies that trace technological evolution over time,

but focussing on a single variable, or methods based on multi-dimensional functions which do not

measure technical position of actors (Lancaster, 1971; Saviotti and Metcalfe, 1984; Sahal, 1985;

Dodson, 1985; Martino, 1985; Tushman and Anderson, 1986, 1990; Trajtenberg, 1990). This separation

is unfortunate, for various reasons.

On a general level, the implicit assumption seems to be that, while in the short run various

technological parameters may influence the survival of a product in the market, with many subtle trade-

offs, in the long run there is a dominant technological dimension that drives the evolutionary dynamics.

Thus computers may differ for a number of characteristics, but in the long run computing power and

cost per computing operation are the only things that matter and dictate the intensity of selective

pressures. Aircraft may be largely different but in the long run direct operating costs (DOCs) and the

number of available seat miles (ASMs) drive the selective success, and so on. In other words, the trade-

offs between parameters can be interpreted as random noise around a trend strictly determined by one

or a few fundamental dimensions.

This claim seems reasonable, but its empirical validity is still to be demonstrated. Our data show

that the dynamics of single technological parameters may deeply differ from the dynamics of an

aggregate measure of performance. More importantly, our data show that rates of change of DEA

measures are both larger and smaller than rates of change in individual parameters. This means that any

conclusion on the nature of technological evolution based on a single indicator is affected by a severe

selection bias.

From a substantial point of view, the reason behind the difference between dynamics in single

measures and dynamics in aggregate multidimensional measures must be explored. In our case,

technology has a strong systemic nature. Technological interdependencies and design

11

complementarities are immensely important. Breakthrough innovations may occur in single

components of a product, producing a big improvement in single technical measures, and in the overall

system. However, in many systemic products, there is a trade-off in the behaviour of single parts of the

system. In this case, a technological discontinuity occurs only when there is innovation at the overall

system level. Complemetarities and trade-offs among parts of the system do not allow to consider

separately innovation. In fact, once a technical advancement occurs in a system, other complementary

technologies must improve for getting a major breakthrough innovation affecting the overall

performance of a product.

4.2 Convergence to the frontier

The comparison between the maximum value of DEA indicators at the level of industry and at the

level of individual competitors is particularly informative (Figure 6). Looking at the relative position of

each manufacturer with respect to the frontier of peak performance, a clear pattern of convergence

emerges.

Although technical progress has been sustained over more than four decades and marked by

significant jumps, all competitors converge towards the technological frontier with their top performing products.

The identity of the technological leader may change over time, but the delay with which followers join

the leader on the frontier is very small. No competitor stays persistently below the frontier with all its

products.

In the first period of the history, Pratt & Whitney and Rolls Royce compete head-to-head and

become technological leaders almost at alternate years. There is no technological dominance of a

competitor over the other. The innovating entry of General Electric in 1970 was soon followed by the

other competitors, which approached the frontier and joined the leader. In the last decade, there is

substitution of technological leadership over time, and all major players converge on the frontier in the

period 1995-1997.

4.3 Survival of unfit products

At the same time, a significant portion of products launched by all competitors lies below the

frontier. Inspection of Figure 7a,b gives an impressive view of the dispersion of individual products in

the technical efficiency space.

At any time, we see the coexistence of two seemingly contradictory processes. On one hand, all

competitors thrive to lead the technical evolution or to join the leader within a short time frame; on the

other hand, competitors go on launching deliberately products that lie below the maximum performance

they can reach in the same period.

12

Figures 8-10 offer a complete view of this phenomenon at the company level. We see two

paradigmatic cases.

Survival by multiplicity of fitness criteria

This case includes two examples. First, after launching the Spey family in the mid-1960s, Rolls

Royce develops several small engines: the M45H and the Tay family (Tay 620, 650 and 651) (see Figure

8). All these models lie well below the frontier of their respective periods, and also below the efficiency

levels that Rolls Royce itself was able to achieve with the almost contemporary RB211 family. In this

case Rolls Royce was induced to develop these models by the need to get through the US market and

by the deliberate choice to serve small customers. In fact, the Tay 651 was introduced in 1980 to enter

into the lucrative multi-source B727 market, while M45H equipped the unsuccessful VFW aircraft, and

Tay 620 and 650 were developed for the Fokker 70 and 100, respectively.

None of these efforts were commercially successful. After the early technological achievements of

the Avon and the Conway, Rolls Royce experienced an immensely difficult period. The Spey family did

not break through the large US market, but was mainly sold to European aircraft manufacturers

(Hawker Siddeley, BAC, Fokker), largely reducing the financial resources available for further

developments. Shortly after Rolls Royce decided to commit itself to a very large engine, anticipating the

growth of the wide-body market. Unfortunately, it targeted the most unsuccessful aircraft in the cut-

throat competition between the Boeing 747, the McDonnell Douglas DC-10 and the Lockeed L-1011

Tristar. As it is well known, the parallel development of the L-1011 and the RB211 placed the existence

of both companies at risk (Newhouse, 1981). Notwithstanding these difficulties, Rolls Royce decided to

respond to the needs of the small engine segment with the Tay family, even though they were not

based on significant advancements.

A slightly different pattern emerges with respect to the CFM56-7 version of the large CFM family,

introduced by the General Electric-Snecma joint venture. As it is clear from (Figure 9), the 7B is the

last but also the least efficient version of a large product family, that also happened to be commercially

highly successful. This version was developed in order to serve the needs of the new generations of the

successful B737 (series 600, 700 and 800). Technically speaking, the 7B version is less performing in

terms of overall pressure ratio and fuel consumption.

These two patterns have an element in common. Although in both cases the groups of products lie

largely below the frontier, they survive thanks to the existence of technological niches. In these niches

users have idiosyncratic functional requirements and value the excellence in a particular technological

dimension more than the overall performance.

13

Note that niches do not imply separate technological clusters or market segments. The examples of

sub-optimal groups of products discussed above are integral part of their respective cluster and

compete with other products in the same segment. While they are less efficient globally, they are

superior to competing products with respect to at least one dimension, which turned out to be decisive

for at least one group of users.

There is a difference between the two cases: in the Rolls Royce case, the overall performance of the

Tay family stays at the same level of the Spey family; in the CFM, on the contrary, the 7B version is

worse. In other words, in one case we have a family extension at the same level of efficiency, by

changing the combination of parameters, in the other a downgrading of the overall performance.

We might think of these niches as formed by users that do not have additive fitness functions, that

is, they have multiple fitness functions. If the benefits associated to the different dimensions of

technological performance of products cannot be expressed through a single measure, then it is

necessary to assume that multiple functions are at place simultaneously. Suppose for example that a

user group has lexicographic selection criteria: inefficient products may survive if they are superior in at

least one dimension, if that matches the most important selection criteria in that group.

Note also that the DEA measure is probably the best approximation to the overall concept of

fitness, because it incorporates all relevant inputs and performance outputs, without imposing any

functional specification to their relationship. In some sense, DEA is the most abstract representation of

fitness in the technological field. The acid test for the modelling strategy based on a single fitness

function is to pass a test against empirical evidence that do not pose severe requirements in terms of

theoretical assumptions on the aggregation of several dimensions of fitness.

Still, as we shall see, in our case a DEA-based measure clearly shows that a single fitness function

does not capture all the technological evolution.

Survival by multiplicity of selection levels

There is a third pattern that merits a careful analysis. Inspection of the Pratt & Whitney and

General Electric/Snecma pictures show a number of clouds of products immediately below the

frontier. In the case of Pratt & Whitney (Figure 10), the JT3D fan family, the JT8D, JT9D and PW4000

all exhibit a small number of models at the frontier and several models within a close distance. In the

JT8D case, for example, early models were at the frontier, while subsequent developments in the

following twenty years were all located roughly at the same level of overall technical efficiency. But,

contrary to the Rolls Royce case, basically all models were targeted to variants of highly successful

aircraft models and were developed at low marginal costs, exploiting robust technological platforms.

Thus, while Rolls Royce tried to develop an entire new family with an obsolete technology (and failed),

14

Pratt & Whitney exploited the old family through a large number of low cost, stretched versions, that

were sold in large quantities. The strong technological advantage and the large commercial success

gained with the top performers of the family (such as the JT8D-7B and 8B with DC-8 and DC-9, and

the JT8D-17R) offered an umbrella for other less efficient models, either in absolute terms or with

respect to the top available technology in the period.

The same logic applies, although with less dispersion in the space of technical efficiency, to the

JT9D family developed for the large B767, B747, DC-10 and A300, and for the PW4000 family. In the

CFM case (see Figure 9), the CFM56 family underwent several developments (series 2, 3 and 5), each

targeted to different aircraft models, and again a large cloud of below-the-frontier engines is visible

during the 1990s.

In this case, survival was due to the extension of the product family. More deeply, selection does

not apply to individual products, but to families of products. If strong product families may survive,

they can shield from selective pressure and offer protection to weaker individual products.

Implications for evolutionary theory

Summing up, we see several examples in which companies deliberately introduce products that

perform worse than they could achieve. In evolutionary terms, we observe unfit entities. These entities

survive thanks to two basic mechanisms: multiplicity of fitness criteria or niche creation, and family

extension.

These two mechanisms refer to fundamental problems in the evolutionary theory of technology.

The possibility of survival of inefficient products in relatively segregated niches raises the problem

of multiplicity of selection criteria or fitness functions. Stated differently, it raises the problem of non-

linearity in the selection environment.

The possibility of survival of inefficient members of an overall efficient product family raises the

problem of the levels of selection. While genic interpretations of evolution (à la Dawkins) admit only one

level of selection, more recent developments make clear that there may legitimately be several levels:

for example, genes, organisms, and communities. As Jablonska (2001) states, it is also possible that the

sign of selective pressures is not the same across levels, so that a give trait is evolutionary favourable at

one level but negative at another.

Our reconstruction of the evolution of jet engine technology is not compatible with a narrow

evolutionary view, which we might label vetero-Darwininian (or à la Dawkins). In this version artefacts

mutate randomly at the level of individual technological parameters and are selected on the basis of a

fitness function that maps the ensemble of parameters on a real number. There are two important

additional assumptions: there is one fitness function, and the fitness function is completely exogenous to

15

the evolutionary process. If this is true, then abstract properties of the evolutionary process of

technologies may be derived from simulation models (e.g. genetic algorithm, NK model) that use a

single fitness function and define it exogenously.

As we have discussed elsewhere there is no compelling reason to accept such a vision, other than

the need to closely follow a strict analogy with the fundamentalist version of evolutionary theory.

In reality, as the contributions in Ziman (2001) clearly show, there are many features in biological

evolution which really encourage a completely different modelling strategy. Among other things, it is

possible to think of an evolutionary process that is based on multiple and endogenous fitness functions

(Jablonski, 2001; Mokyr, 2001). Allowing for multiple fitness functions implies the existence of several,

irreducible selection criteria, with no need to reconcile them through a common metrics. Allowing for

the endogeneity of the fitness function means that the outcome of the selection process may have a

feedback on selection criteria, so that these change over time.

As a matter of fact, our story calls for a more “liberal” evolutionary interpretation, in which directed

mutation is allowed, there are several levels of selection, and selection follows a number of non

reducible criteria.

4.4 Structural evolution and discretionary strategic behaviour

One of the crucial points in the debate on evolutionary modelling of technology is whether a

structural description captures all the stylised facts or there is some need to incorporate strategic

subjectivity. Stated differently, there is debate over the limits to strategic manoeuvring of companies set

by structural characteristics of the dynamics of technology, such as paradigms, trajectories, path

dependency, inertia and the like (Dosi, Nelson and Winter, 2000).

At one extreme, pure structural models resolve strategic discretionary behaviour as a collection of

meta-routines, which govern more or less complex articulations of routines at various levels. The

choice of strategy is severely limited by the available repertoire, which in turn is a function of existing

competencies. Although various “learning to learn”, meta-levels and abstract or general competencies

are introduced, the basic idea is that the intrinsic features of technological learning place severe

limitations to the degrees of freedom of companies.

Our detailed reconstruction offers some valuable insights on this debate.

More precisely, we offer an articulation of the relationship between structural technological

dynamics and the range of company technological strategies that are compatible with that dynamics.

There are several structural features of evolutionary dynamics of technology in the jet engine industry,

namely:

16

- the overall evolution is characterised by three technological jumps, followed by incremental

developments;

- there is convergence on the technological frontier by all competitors, at least with their

respective top performer products;

- at the same time, there is persistence of sub-optimal products over time, due to the creation of

niches and the extension of product families;

- all competitors develop large families of products after having reached peak levels of

technological performance;

- entry takes place only in structural holes (see CF6 for General Electric, the CFM family, and the

entry of Allison).

There are also variations in the way different competitors place themselves within this structural

dynamics and, in turn, contribute to shape it, namely:

- Pratt & Whitney and Rolls Royce follow a strategy of sequential entry into different market

segments, while General Electric plays a game of late entry into all segments simultaneously;

- in the head-to-head competition between P&W and R-R, the former is more rapid in

introducing radically new engines and more aggressive in the gradient of change within its

product portfolio, while the latter is laggard and more gradual;

- General Electric enters, both in isolation and in alliance with Snecma, with highly superior

products from the technological point of view, displacing competitors with aggressive offers;

- the dynamics of introduction of new products is one of point-to-point, cut-throat competition;

- there is an increase over time of the extent of progress allowed by successive platforms: from

0.2 to 0.3 for Spey and Tay in the Rolls Royce case and for JT4 and JT8 for Pratt & Whitney;

from 0.6 to 0.8 for RB211 and from 0.7 to 0.9 for CFM56;

- there is also an increase in the length of life of product families, from a few years for Avon,

Spey and Tay to many years for the JT3, JT8 and JT9 and for the CFM56;

- the last two trends seem to imply than competitors try to endogenise technological

discontinuities, optimising the exploration-exploitation trade-off and making successive

technological jumps less disruptive.

Summing up, the overall technological dynamics place severe constraints to the set of feasible

strategies. The steady but discontinuous evolution of technologies require a steady flow of investment

and the ability to balance inter-temporally the huge financial needs and the cash flows. In order to

survive, companies must be able to stay at the technological frontier, at least with part of their product

portfolio. All companies must solve in some way an extremely severe exploitation-exploration trade-

off.

17

At the same time, there are some degrees of freedom for variety in strategic approaches. Both

leaders and followers can gain market shares. Companies may optimise their product portfolio in

different ways in terms of timing and extent of investment. Companies may also play different

strategies in terms of the length of product families. Strategic options are constrained but not

eliminated by structural features of technological evolution.

No purely structural representation is able to capture the subtle variety of strategic approaches. In

turn, the implementation of these alternative approaches may shape the overall dynamics differently.

Interestingly, the internalisation of discontinuities that manufacturers pursue through the lengthening

of life of families and the robustness of technological platforms have a structural effect, insofar as

successive jumps are less radical. The three discontinuities, in fact, involve progressively lower short

term rates of change.

Thus we have a feedback effect, from company strategies to the structural evolution. This

represents another important contribution to the debate on evolutionary modelling.

4.5 Technological performance and market success

An evolutionary account of technology must acknowledge, ultimately, that selection takes place not

on the basis of technological performance alone, but also on the basis of economic considerations. The

history of technology and of economic growth is rich of examples where fully developed technologies

did not diffuse in the economy until their economics, perhaps due to exogenous factors such as relative

prices or raw materials, became attractive (Rosenberg and Birdzell, 1990; Mokyr, 1993; Landes, 1999).

Nelson (2001) has convincingly argued that the articulation between selection by the community of

technologists based on technical performance, and selection by the community of users based on

economic factors is one of the crucial points for a theory of evolution of technology. Although this

theme is considered critical, a few reconstruction and detailed stories are available.

We may explore the detailed relationship between technological performance and market success

across all the history of the industry.

To carry out this analysis, we preliminarily plotted for each engine the value of DEA index against

one indicator of market success: the yearly average number of products sold during its life. In order to

make DEA measures and market shares comparable across time, we did not consider in this

preliminary exploration all engines introduce since 1990. The plot of data is available in Figure 11.

It is clear from data that there is not a close relation between technological superiority and market

success. Not only technological performance and economic value are not synonymous, they are also

uncorrelated. If we observe separatedly the three jumps of innovation according to levels of DEA, we

cannot detect a positive correlation between the two variables. On the contrary we observe high selling

18

products which in many cases have a level of efficiency lower than the other engines in the same group.

This result deserves further analysis and statistical testing to be made more robust and general.

If confirmed, this result has a profound implication on evolutionary modelling. First, we have

already showed that, even within the narrow region of technology, a single fitness function and a single

selection level do not represent the evolutionary process correctly. To this we now add a new

dimension: a comprehensive representation must admit that a new, non reducible, level of selection

must be introduced and that a new type of selection criteria are at work in the evolution of

technologies.

Technologies evolve following a variety of internal criteria, that are not reducible to any single

dimension, and following an endogenous economic dynamics mediated by the market.

The exact relationships between technological performance and market success must be the object

of a future detailed study.

References

Anderson J. D., 1989, Introduction to flight, McGraw-Hill International Edition.

Anderson P., Tushman M. L., 1990, Technological discontinuities and dominant designs: a cyclical model of

technological change, Administrative science quartely, 35, 604-633.

Barnard R. H., Philpott D. R.,1995, A description of the physical principles of aircraft flight, Logman.

Bonaccorsi A., Giuri P. 2001a, Technological competition and network structure in the aero-engine industry,

LEM Working Paper 2001, Sant'Anna School of Advanced Studies, Pisa, prepared for the Nelson and

Winter Conference in Aalborg, Denmark June 12-15 2001.

Bonaccorsi A., Giuri P., 2001b, The long term evolution of-vertically related industries, International Journal of

Industrial Organization, 19, 1053-1083.

Bonaccorsi A., Giuri P., 2001c, Network Structure and Industrial Dynamics. The long term evolution of the

aircraft-engine industry, Structural Change and Economic Dynamics, 12, 201-233.

Charnes A., Cooper W.W.,1985, Preface to topics in data envelopment analysis, Annals of Operations Research,

2, 59-94.

Constant E.W. 1980, The origin of the turbojet revolution, The John Hopkins University Press, Baltimore and

London.

Dodson E.N., 1985, Measurement of state of the art and technological advance, Technological Forecasting and

Social Change, 27, 129-146.

Dogramaci A., Färe R., 1988, Applications of modern production theory: efficiency and productivity, Kluwer

Academic Publishers.

19

Dosi G., Nelson R. R., Winter S.G., 2000, The nature and dynamics of organizational capabilities, Oxford

University Press.

Färe R., Grosskopf S., 2000, Reference guide to OnFront, Economic Measurement and Quality Corporation.

Farrell M. J., 1957, The measurement of productivity efficiency, Journal of the Royal Statistical Society, ser.

A,General, 120, 253-281.

Fried H., Knox Lovell C.A., Schmidt P., 1993, The measurement of productive efficiency-Techniques and

applications, Oxford University Press.

Garvin, 1998

Greene W. H., 1993, The econometric approach to efficiency analysis in Fried H., Knox Lovell C.A., Schmidt P.,

1993, The measurement of productive efficiency - Techniques and applications, Oxford University Press.

Griliches Z., 1971, Price indexes and quality change, studies in new methods of measurement, Price Statistics

Committee Federal Reserve Board, Cambridge, Harvard University Press.

Grosskopf S., 1993, Efficiency and productivity in Fried H., Knox Lovell C.A., Schmidt P., 1993, The

measurement of productive efficiency - Techniques and applications, Oxford University Press.

Grupp H. 1998, Foundations of the economics of innovation: Theory, measurement and practice, Edward

Elgar.

Jablonka E., 2001, Lamarckian inheritance systems in biology: a source of metaphors and models in

technological evolution, in Ziman J., 2001, Technological innovation as an evolutionary process,

Cambridge University Press.

Jane’s Aero-engines, 1997, Jane’s Information Group, Sentinel House.

Jane’s All the Worlds Aircraft (1940-1998), Jane’s Information Group, Sentinel House.

Knox Lovell C.A., Schmidt P., 1988, A comparison of alternative approaches to the measurement of productive

efficiency, in Dogramaci A., Färe R., 1988, Applications of modern production theory: efficiency and

productivity, Kluwer Academic Publishers.

Lancaster K.J., 1971; Consumer demand, a new approach, New York, London.

Levinthal D. A., 1998, The slow pace of rapid technological change: gradualism and punctuation in technological

change, Industrial and corporate change, 7, 217-247.

Martino J.P., 1985, Measurement of technology using tradeoff surfaces, Technological forecasting and social

change, 27, 147-160.

Miller R., Sawers D. (1968), The Technical Development of Modern Aviation, Routledge & Kegan Paul,

London.

Mokyr J., 1993, Evolutionary phenomena in technological change in Ziman J., 2001, Technological innovation as

an evolutionary process, Cambridge University Press.

20

Mowery D.C., Rosenberg N. (1982), “The Commercial Aircraft Industry”, in Nelson R. (ed.), Government and

Technological Progress, Pergamon Press, New York.

Nelson R., 2001, Selection criteria and selection processes in cultural evolution theories in Ziman J., 2001,

Technological innovation as an evolutionary process, Cambridge University Press.

Norman M., Stoker B. (1991) Data Envelopment Analysis- The assessment of performance, John Wiley and

sons

Phillips A. (1971), Technology and Market Structure. A study of the Aircraft Industry, Heath Lexington Books, Lexington,

Massachusetts.

Sahal D., 1985, Technological guideposts and innovation avenues, Research Policy, 14, 61-82.

Sahal D., 1985b, Foundation on technometrics, Technological Forecasting and Social Change, 27, 1-37.

Saviotti P.P., Metcalfe J.S., 1984, A theoretical approach to the construction of technological output indicators,

Research Policy, 13, 141-151.

Trajtenberg M, 1990, Economic analysis of product innovation - The case of CT scanners, Cambridge, Harvard

University Press.

Tushman M. L., Anderson P., 1986, Technological discontinuities and organizational environments,

Administrative science quartely, 31, 439-465.

Vincenti W.G. (1990), What Engineers Know and How They Know It. Analytical Studies from Aeronautical

History, The John Hopkins University Press, London.

Ziman J., 2001, Technological innovation as an evolutionary process, Cambridge University Press.

21

Figure 1 Max DEA efficiency at the firm level

Figure 2 Max DEA efficiency at the firm level, cluster 1

0

0,1

0,2

0,3

0,4

0,5

0,6

0,7

0,8

0,9

1

1956

1958

1960

1962

1964

1966

1968

1970

1972

1974

1976

1978

1980

1982

1984

1986

1988

1990

1992

1994

1996

1998

AECFMGEIAPWRRTX

RR AVON

RR SPEYRR TAY

GE CF6

RR RB211

GE CF34

GE90RR TRENT

RR CONWAY

PW JT3DPW JT8

PW JT9

CFM56

PW4000

IAV2500

PW2000

0

0,1

0,2

0,3

0,4

0,5

0,6

0,7

0,8

0,9

1

1956

1958

1960

1962

1964

1966

1968

1970

1972

1974

1976

1978

1980

1982

1984

1986

1988

1990

1992

1994

1996

1998

AECIGEIAPWRRTX

RR CONWAY

RR SPEYRR TAY

PW JT3DPW JT8

CFM56 GE CF34

IAV2500

TX ALF

AE 3007

RR AVON

22

Figure 3 Max DEA efficiency at the firm level, cluster 2

Figure 4 Max DEA efficiency at the firm level, cluster 3

0

0,1

0,2

0,3

0,4

0,5

0,6

0,7

0,8

0,9

119

56

1958

1960

1962

1964

1966

1968

1970

1972

1974

1976

1978

1980

1982

1984

1986

1988

1990

1992

1994

1996

1998

GEPWRR

RR RB211

PW4000

PW2000

PW JTD9D

GE CF6

0

0,1

0,2

0,3

0,4

0,5

0,6

0,7

0,8

0,9

1

1956

1958

1960

1962

1964

1966

1968

1970

1972

1974

1976

1978

1980

1982

1984

1986

1988

1990

1992

1994

1996

1998

GEPWRR

GE CF6

RR TRENT

GE 90

PW 4000

23

Figure 5 3-years rate of change

Figure 6 DEA efficiency, technological frontier of larger firms

1

1,5

2

2,5

3

3,5

4

4,5

5

1961

1964

1967

1970

1973

1976

1979

1982

1985

1988

1991

1994

1997

BPRDEAOPRT/WSFC

0,1

0,2

0,3

0,4

0,5

0,6

0,7

0,8

0,9

1

1958

1960

1962

1964

1966

1968

1970

1972

1974

1976

1978

1980

1982

1984

1986

1988

1990

1992

1994

1996

CFMGEIAPWRR

PW

GE

IACFM

RR

24

Figure 7a DEA efficiency at the product level, 1

199019881986198419821980197819761974197219701968196619641962196019581956

,5

,4

,3

,2

,1

0,0

TAY

SPEY

M45H

JT8

JT4

JT-3

Conway

AVON

SPEY 555SPEY 555 JT8D-17R

TAY 620-TAY 651- TAY 650-M45H

JT8D-219JT8D-217SPEY 506

SPEY 1JT8D-217JT8D-217

SPEY 512SPEY 511JT8D-17AJT8D-15AJT3D-7

JT3D-3DJT3D-3CJT3D-3BJT8D-9AJT8D-9JT8D-7B JT8D-7AJT8D-7 JT8D-17JT8D-15JT8D-11 JT8D

CONWAY 5CONWAY 5

AVON 533AVON 531AVON 527JT3C-7

JT4A

JT3C

25

Figure 7b DEA efficiency at the product level, 2

19981996199419921990198819861984198219801978197619741972197019681966

1,05

1,00

,95

,90

,85

,80

,75

,70

,65

,60

,55

,50

V2500

Trent

RB211

JT9

CFM56

CF6

CF34

ALF

90

4000

3007

2000

Trent 89Trent 88407490-92B

Trent 87Trent 77CFM56-5CCFM56-5C

Trent 87

Trent 76

4077CFM56-5CCFM56-5B

RB211-52 CFM56-5BCFM56-5A

CF6-80E1 CF34-3B1CFM56-5BCFM56-5ACF6-80C2 CFM56-5A

2037RB211-52 4090CF34-3A1

2040 CFM56-5B

CF6-6D4360

JT9D-7R4 4460JT9D-7R4 4056 CFM56-5BCFM56-2

V2500CFM56-2C4158CF6-80A2 V2533-A5V2530-A5

4156JT9D-3A 4152 4060CF6-80ACFM56-3C

RB211-53CF6-50E2CF6-50C2CF6-50C 41684164CFM56-3BRB211-52JT9D-7JJT9D-7AJT9D-7 CFM56-7B

JT9D-7QJT9D-7FV2528

RB211-22ALF507-1ALF507-1ALF502-R V2527V2525 V2522

RB211-52RB211-52JT9D-59A CF6-45A2

RB211-53

3007A

26

Figure 8 DEA efficiency at the product level, Rolls Royce

1998199619941992199019881986198419821980197819761974197219701968196619641962196019581956

1,1

1,0

,9

,8

,7

,6

,5

,4

,3

,2

,1

RR- Trent

RR- TAY

RR- SPEY

RR- RB211

RR- M45H

RR- Conway

RR- AVON

RB211-53

RB211-53

RB211-52

RB211-52

RB211-52

RB211-52RB211-52RB211-22

Trent 89Trent 88

Trent 87Trent 87

Trent 77

Trent 76

TAY 651- TAY 650-TAY 620-

SPEY 555SPEY 555

SPEY 512SPEY 511

SPEY 506SPEY 1

M45H

CONWAY 5CONWAY 5

AVON 533AVON 531AVON 527

27

Figure 9 DEA efficiency at the product level, CFM International and General Electric

199819961994199219901988198619841982198019781976197419721970

1,1

1,0

,9

,8

,7

,6

CI- CFM56

GE- CF6

GE- CF34

GE 90

CF6-80A2

CF6-80A

CF6-6D

CF6-50E2CF6-50C2CF6-50C

CF6-45A2

CF6-80E1CF6-80C2

90-92B

CF34-3B1

CF34-3A1

CFM56-7B

CFM56-5C

CFM56-5CCFM56-5C

CFM56-5B

CFM56-5B

CFM56-5B

CFM56-5B

CFM56-5BCFM56-5ACFM56-5A

CFM56-5A

CFM56-3CCFM56-3B

CFM56-2CCFM56-2

28

Figure 10 DEA efficiency at the product level, Pratt & Whitney

1998199619941992199019881986198419821980197819761974197219701968196619641962196019581956

1,1

1,0

,9

,8

,7

,6

,5

,4

,3

,2

,1

PW- JT9

PW- JT8

PW- JT4

PW- JT-3

PW- 4000

PW- 2000

JT9D-7R4JT9D-7R4

JT9D-7QJT9D-7JJT9D-7FJT9D-7AJT9D-7

JT9D-59A

JT9D-3A

4460

4158 41564152 4060

4056

20402037

4360

41684164

4090

4077

4074

JT8D-9AJT8D-9JT8D-7B JT8D-7AJT8D-7

JT8D-219JT8D-217

JT8D-217JT8D-217

JT8D-17R

JT8D-17AJT8D-17

JT8D-15AJT8D-15JT8D-11 JT8D

JT4A

JT3D-7JT3D-3DJT3D-3CJT3D-3B

JT3C-7

JT3C

29

Figure 11 Technical efficiency vs market success

yearly average product sales

240220200180160140120100806040200

DEA

effi

cien

cy1,0

,8

,6

,4

,2

0,0

TX

RR

PW

IA

GE

CI

TAY 650-15

4060V2500

RB211-524G

4056

TAY 620-15

4152

2040CFM56-5A1

4158JT9D-7R4E

2037

CFM56-3B

RB211-535C

JT8D-219

RB211-535E

JT8D-217A

CF6-80A2CF6-80A

ALF502-R5

JT8D-217

TAY 651-54

RB211-524CCF6-45A2

RB211-524DCFM56-3C

JT9D-7QCF6-50E2

JT8DJT8D-17AJT8D-15A

CFM56-2C1

SPEY 555-15H

JT8D-217C

CF6-80C2

CF6-50C2

RB211-524BRB211-22B

M45H

JT8D-17

CF6-50C

CFM56-2 JT9D-7R4

JT9D-7JJT9D-7F

CF6-6D

JT9D-7JT9D-3A

JT8D-17R

JT9D-7A

JT9D-59A

JT8D-9AJT8D-9JT8D-11

SPEY 555-15

JT8D-7A JT8D-15JT3D-7SPEY 512SPEY 511

SPEY 506SPEY 1

JT8D-7BJT8D-7

AVON 533RCONWAY 509CONWAY 508

AVON 531B

JT3D-3D

JT3C-7AVON 527 JT4A

JT3D-3C

JT3C

JT3D-3B